In recent years, artificial intelligence (AI) applications have accelerated, and AI customer service has been introduced by many platforms due to its low cost and fast response. Consumers do not need to queue up. They only need to click on the dialog box to inquire about returns and exchanges, check logistics, and handle returns and changes at any time. This brings convenience to both enterprises and users, and also relieves the pressure of manual customer service to a certain extent.
However, the rapid popularity of AI customer service has also exposed a series of worrying problems. Some people's express delivery has been stalled for many days, and the customer service phone has never been answered; some people bought movie tickets by mistake, and missed the refund time because they could not find the manual entrance in the APP; some people repeatedly entered "transfer to manual", but all they got was the AI sent "I understand how you feel."
Complaint data confirms this pain point. In the "Analysis of Complaints Accepted by National Consumers Association Organizations in the First Half of 2026" released by the China Consumers Association in August this year, complaints against AI customer service are particularly prominent, focusing on its false promises, difficulties in manual access, and inaccurate generated content. Annual data previously released by the Black Cat Complaint Platform also showed that among complaints in the e-commerce field in 2025, customer service attitude problems accounted for more than 40%, and many consumers spent a lot of time and energy on "switching to manual work."
Faced with the widespread application of intelligent customer service, how to find a balance between reducing corporate costs and protecting consumer service rights has become a top priority.
Looking closely at these problems, the crux is not the technology itself, but the application method. The introduction of AI customer service is a reasonable choice to reduce costs and increase efficiency: simple, high-frequency, standardized questions are answered by machines, consumers receive immediate responses, and enterprises release manpower pressure. This is understandable. The key lies in grasping the degree. For example, some companies put their customer service assessment on the "AI interception rate", and the manual entrance is gradually postponed; the customer service, order, and work order systems are independent of each other, and consumers can only "start from the beginning" when switching between humans and machines.
Therefore, switching to manual difficulties, switching gears, and breaking promises are more than just experience flaws. Article 59 of the "E-Commerce Law" requires platforms to establish a convenient and effective complaint and reporting mechanism. "Convenient" means that the entrance can be found and accessible; "effective" means that problems can be corrected and losses can be stopped. When technical settings objectively raise the threshold for consumers to obtain services, the platform’s service obligations are compromised.
In response to this pain point, national-level regulations have emerged.
On September 1, the "Customer Contact Service - Requirements for Collaboration of Artificial and Intelligent Customer Service" (GB/T 47746-2026, hereinafter referred to as the "Requirements") was officially implemented, providing specifications on how artificial and intelligent customer service collaborate, when they must switch to manual, and how to connect information when switching. AI customer service has an "instruction manual".
This standard provides a ruler to the industry, measuring the boundaries between efficiency and responsibility: the customer service entrance must be clear and eye-catching, for example, "Customer Service" can be marked in large characters at the top of the website homepage; manual transfer must support multiple triggering methods such as keywords, menus, buttons, and voice; intelligent customer service If the interaction fails to reach the set threshold, the customer's negative emotions are sensed, or there is an emergency situation involving personal and property safety, the system should automatically transfer to manual labor; when switching between humans and machines, the customer's identity, historical interaction records, and problem descriptions must be transmitted together. It is not appropriate to ask the customer to repeat the stated problems after the switch.
As a technical requirement that plays a leading role in the industry, the "Requirements" will raise the bottom line of AI customer service collaboration and force manual takeover and information connection to become service standards.
At the administrative level, market supervision departments can use it as a reference to judge whether there are defects in corporate services, and then take measures such as interviews, supervision of rectification, or mediation. When a service dispute occurs, the court may also use it as a reference for judging common industry practices. In addition, if the standard is adopted by industry authorities as the basis for service quality evaluation, it will also drive companies to actively follow it and become a market driving force for companies to actively benchmark.
Promoting the implementation of the "Requirements" cannot be separated from the joint efforts of regulatory authorities, enterprises and consumers. Regulatory authorities transform national standards into a basis for evaluating service quality and mediating disputes, which can guide companies to make improvements; companies proactively align with standards and internalize the "Requirements" as a conscious effort to upgrade services, which can make entrances brighter, handovers smoother, and promises more substantial; consumers who encounter obstacles in transferring to labor and failed promises can keep screenshots and recordings and make rational complaints - every invocation of standards is an external motivation to promote corporate service upgrades.
After all, AI customer service does not need a job number, but the responsibility must have a subject; the machine can run the process, but the promise must be claimed by someone. "Switching to manual labor" should not be a step into a maze, but the beginning of high-quality services.
